Both M-ary amplitude-shift keying (ASK) and M-ary quadrature-amplitude modulation (QAM) have been widely used in digital communications, where M is the signal constellation size. In digital modulation with an M-ary constellation, each block of k=log2M bits are mapped into an M-ary symbol, where k is a positive integer. The signal constellation, consisting of M points, is a geometric presentation of the candidate baseband signals to transmit. The M points are distributed on a real axis for ASK or on a complex plane for QAM to represent the M possible real or complex symbols, respectively. In an existing regular ASK or QAM constellation, the M signal points are evenly distributed, that is, the distance between any two neighboring signal points of the constellation is equal. The average bit-error rate (BER) performances, that is, the relationship of BER vs. the signal-to-noise ratio (SNR), of different bits among the k bits are quite similar.
In the application of digital speech communications, the output bits produced by a speech codec are typically reordered into a sequence of descending importance. The most important bits are those that will have the greatest impact upon the received voice quality if they are received in error. Errors to the least significant bits will have only a negligible impact upon the received voice quality. An example of this may be found with the AMR codec used in UMTS cellular systems. See 3GPP TS26.101 section 4.2.1 hereby incorporated by reference in its entirety.
It is therefore usual practice to provide additional error protection to the most important bits. The bits are therefore classed into groups. The first class of bits will receive the greatest protection. The second class will receive relatively less protection, and so on for however many classes of bits there are. An example of this may be found with the AMR codecs used in UMTS cellular systems. See 3GPP TS26.101 section 4.2.2, hereby incorporated by reference in its entirety. In that example, the importance classes are Class A, Class B, and Class C. Class A contains the bits most sensitive to errors and any error in these bits typically results in a corrupted speech frame which should not be decoded without applying appropriate error concealment. This class is protected by the Codec CRC in AMR Auxiliary Information. Classes B and C contain bits where increasing error rates gradually reduce the speech quality, but decoding of an erroneous speech frame is usually possible without annoying artifacts. Class B bits are more sensitive to errors than Class C bits. The importance ordering applies also within the three different classes and there are no significant step-wise changes in subjective importance between neighboring bits at the class borders.
In another example, when the AMR codec is used with a GSM full rate channel, the sequencing and classification of bits from speech codecs is given in TS 45.003 in section 3.9.4., hereby incorporated by reference in its entirety. The protection classes are: 1a—Data protected with the CRC and the convolution code; and 1b—Data protected with the convolution code. No unprotected bits are used.